Benchmarks
Reproducible SERP API benchmarks
Speed, reliability and price, measured the same way for every provider, with open-source scripts you can run yourself. First public run: coming soon.
First public run: coming soon
This page describes the methodology and our targets. There are no measured results here yet, and we won't publish any until the scripts and raw data are public on GitHub.
What we measure
Six numbers that matter for production and agents
Latency p50 / p95
Wall-clock time from request to full response body, measured from a neutral client, per provider, country and endpoint.
Success rate
Share of requests that return a non-empty, correctly parsed result page. Blocks, timeouts, 5xx and empty parses count as failures.
URL resolution
Share of organic links returned as the real destination URL rather than a Google redirect or tracking link.
Price per 1,000
Effective cost of 1,000 successful searches at public list prices, including billing for failed requests where a provider charges for them.
Tokens per result
Estimated LLM tokens for the provider's default response and its most compact LLM-oriented format, where one exists.
How
Methodology
- Fixed query set. A published list of queries across countries and intents (informational, commercial, local, news), frozen before each run.
- Same client for everyone. One open-source runner calls each provider's documented Google web search endpoint with equivalent parameters (country, language, 10 results), from the same region, with the same concurrency.
- Reference fetch. To score parse quality we compare against a reference fetch of the same query and country taken at the same time.
- Monthly runs. Results are published monthly with the raw per-request data (status, latency, parsed fields), so outliers and regressions are visible.
- List prices only. Price per 1,000 uses public list prices at defined volumes, including expiry rules and whether failed requests are billed. No private discounts.
We only benchmark logged-out, publicly visible results, the same way SerpKite collects them. See how we collect data.
Live now
Production latency and success, last hour
Straight from GET /status on our API. This is our own measurement; the independent monthly benchmark is below.
Loading live metrics…
Targets
SerpKite targets
From our product spec. These are commitments we'll measure against, not measured results.
| Metric | Target | Measured |
|---|---|---|
| Latency p50 (real-time /v1/search, US/EU) | ≤ 1.2 s | Coming soon |
| Latency p95 (real-time /v1/search, US/EU) | ≤ 2.5 s | Coming soon |
| Success rate | ≥ 99.0% | Coming soon |
| API uptime | 99.9% | Coming soon |
| URL resolution | Every link decoded to its destination | Coming soon |
| Price per 1,000 | $0.60–$1.00 | Coming soon |
Providers
Who's in the first run
Public list prices for Google web search, $ per 1,000 (verified 2026-09-28). Latency and success columns fill in after the first run.
| Provider | ~$50 tier | ~100k/mo | ~1M/mo | Best list | Credit expiry | p50 / p95 | Success |
|---|---|---|---|---|---|---|---|
| SerpKite | $0.80 | $0.60 | $0.60 | $0.60 | Never | Coming soon | Coming soon |
| Serper | $1.00 | $0.75 | $0.50 | $0.30 | 6 months | Coming soon | Coming soon |
| SerpApi | $15.00 | $7.25 | $3.75 | $1.96 | Monthly reset | Coming soon | Coming soon |
| SearchApi | $4.00 | $2.50 | $1.50 | $1.00 | Monthly | Coming soon | Coming soon |
| Scrapingdog | $1.00 | $0.45 | $0.29 | $0.14 | Monthly | Coming soon | Coming soon |
| Tavily | $7.50 | $5.00 | enterprise | – | Monthly | Coming soon | Coming soon |
| Brave | $5.00 | $5.00 | $5.00 | $5.00 | Monthly | Coming soon | Coming soon |
SerpKite columns use the Growth pack for the ~$50 tier and the $300 Pro pack (the largest self-serve pack) above that; larger volumes are quoted on request. Detailed comparisons: vs Serper, vs SerpApi, best SERP APIs.
FAQ
Benchmark questions
Start building
Measure it yourself
Run the same queries in the playground, or grab a free key and benchmark against your own workload.